Skip to content

Dataset · basis-ai

Multilingual emotional speech dataset

basis-ai/basis-global-emotions

Over 525 hours of audio with expressed emotions and speaking styles across 108 locales for voice synthesis and analysis.

What's inside

The dataset includes 137,696 audio recordings (525.17 hours) from 10,420 native speakers across 108 locales, including Arabic, Amharic, Afrikaans, Azerbaijani, and others. In each recording, the speaker delivers a line with a specified emotion (e.g., joy or irritation) or delivery style (whisper, shout). Recordings are grouped into 54,575 series where the same person reads the same phrase with different intonation.

What you can do with it

  • Training emotional text-to-speech models
  • Audio classification of emotions and intonations in voice
Tasks
Text → speech, Audio classification
Languages
Afrikaans, Akan, Amharic, Arabic, Assamese, Azerbaijani, Bulgarian, Bangla and 4 more
Size
100K–1M rows
Downloads this month
6.5K
Likes
56
Access
on request on Hugging Face

How to load it

from datasets import load_dataset

ds = load_dataset("basis-ai/basis-global-emotions")